[PATCHJ] Fix for diagnosing half pointers/arrays

[John McCall]

On Feb 8, 2013, at 8:32 AM, Joey Gouly <joey.gouly at arm.com> wrote:
> In r173254 I committed some changes to add support for 'half' type to clang.
> 
> I got some of the error cases wrong, and produced a diagnostic for '&h[0]'
> and '&*h',
> which are both defined to always produce a pointer. C99 6.5.3.2, footnote
> 84.
> 
> Does this look like a good fix?

You don't actually care about *why* you have an lvalue of half type;  you just
care that you *do*, and that you're trying to do an lvalue-to-rvalue conversion
on it.  Just check whether the lvalue expression's type isHalfType() instead
of trying to enumerate all the syntactic cases that can give rise to it.

John.